Nicholas Rescher's Coherence Theory of Truth

NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH

Diderik BATENS

1. A ca re fu l re a d e r o f Nich o la s Rescher's Hyp o th e tica l Reasoning mig h t have remarked th at the logical machinery presented in that book has a much wider range of application than the e xp licit discussion o f th e b o o k suggested. I n Th e Coherence Theory of Truth (1 this machinery is not o n ly applied to a wid e r domain o f problems, b u t comes t o p la y a central role in a preeminently philosophical theory, a theory of truth. The label "coherence th e o ry o f tru th " ma y be somewhat misleading a t first sight. The ma in subject o f the book is a criterion and n o t a definition. (Rescher himself defends th e suitability of defining truth in correspondence terms.) Further- more, correspondence considerations a s w e l l a s pragmatic considerations a re also made use o f in Rescher's coherence theory. I shall first t ry to make roughly clear how these three components are interrelated. The problem is to decide, in a given situation, the truth status of some proposition. The starting point is data. Un like in a correspondence theory, these data are not taken as completely certain. They are not truths, but merely truth candidates. The coherence machinery is applied t o th e data, and o n ly as a result of this process does one get truths. Pragmatic considerations come up in the justification of the whole procedure. Not- withstanding the important role o f correspondence and pragmatic considerations, th e crite rio n is essentially coherentist in that (a) the data as such guarantee no truth at all, and may indeed be fla tly inconsistent, leading to truth o n ly after they have been processed by the coherence machinery, and (b) the pragmatic considerations are used only in support of the valid- ity of the whole procedure, and not in support o f particular truth-claims.

( Press,1 1973. ) N i c h o l a s R E S C H E R : T h e c o h e r e n c e t h e o r y o f t r u t h . O x f o r d , C l a r e n d o n 394 DIDERIK BATENS

It goes without saying that the objection that "a set of false- hoods may be as coherent as a set of truths" is pointless with respect t o a th e o ry which, a s the one under consideration, starts fro m experiential data. Of course, one might repeat the objection in a more sophisticated way, and ask wh y the application of the coherence machinery to the data should lead to a set o f true propositions. This objection can be directed against th e specific coherence machinery presented i n t h e book under discussion, o r against a n y coherence machinery in general. I shall return to the first version o f the objection in the sequel, and briefly consider the second version here. The concept of ''truth" that the objector has in mind cannot be the coherence concept o f tru th itself; indeed, e ve ry reasonable criterion of truth would accord with his own truths for trivia l reasons. What the objector has in mind is probably the defini- tional concept of truth as correspondence with the facts. The objection has then to do with the fa llib ility of the coherentist criterion, and not o n ly wit h its fa llib ility wit h respect to its own different applications i n time, b u t wit h respect t o th e facts as well. Tarski's definition of truth is unexceptionable of course, not only for the trivia l reason that it can be read as a (stipulative) definition, b u t also because i t expresses with o u t a n y doubt exactly what we mean b y 'true'. But precisely because it is infallible, it is useless as a criterion o r even as a basis fo r a criterion. Indeed, a n y criterion o f truth will be either almost application-free (lead to an empty set of truths) o r fallible, and this not for a theoretical reason, but for the simple and contingent reason that our experiential means are as they are. Con- sequently, a ll one can require from a criterion of truth is that it lead to 'p is true o n ly if there is a rational warrant for 'p is true'. This rational warrant cannot be found in correspondence considerations f o r reasons mentioned before (protocol sentences are a myth). Nor can it be found immediately in pragmatic considerations, for they lead at best to useful tools, and, as Karl Popper argued in his Logic of Scientific Discovery, tools cannot be true o r false. This suggests that a rational warrant has to depend at least to some extent on coherence considera- NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 395 lions. O n e ca n argue against either th e re la tive imp o rt o f coherence considerations in a criterion of truth, or the specific coherence machinery that is suitable, but the applicability of coherence considerations in general mu st be acknowledged. Returning to the question of whether any coherence criterion will guarantee the truth of its results, the answer seems to be that it never will guarantee truth in an absolute sense, but that some such theory will account for a rational warrant for truth, if any theory at all will. Of course, it never will do so on its Own; without data it is completely worthless. And there surely is no a priori way to find out which specific coherence criterion is to be adopted, nor even to what extent coherentist considerations have to play a role. But given the kind of epistemic subjects we think ourselves to be, and given the incomplete reliability o f o u r techniques f o r gathering data, a tru th criterion without any coherence machinery can hardly be defended. The fact that there is fa llib ility, and the fact that there is rational warrant only, might suggest that we are dealing with a theory of acceptance rather than with a theory of truth. Wha t is meant here, however, is n o t a th e o ry o f acceptance lik e many o f those n o w extant, wh ich account o n ly f o r general hypotheses and predictions, including so called retrodictions. Most of these theories start comfortably from a set of unques- tioned "evidence." However, a criterion of truth has to account even fo r this "evidence," o r expressed more suitably, f o r the factual truths that are used as a basis for accepting predictions and general hypotheses. Supposed that we use 'theory of acceptance' in the broader sense, what are the reasons wh y the coherence theory cannot be considered as such a theory ? As Rescher argues, the rational warrant "applies to the criterion at issue, and not (save obliquely) with the propositions to which that criterion addresses itself ..." (p. 16). Two other differences seem relevant here. If an acceptance theory has to be a useful explicatum f o r wh a t happens, e.g., i n th e actual h isto ry o f science, it seems that it has to allow, a t least under definite conditions (a) the acceptance of inconsistent theories, and (b) the acceptance at the same time of some theory and of falsifying instances o r even a falsifying lo we r le ve l (read less 396 DIDER1K BATENS general) hypothesis. Such points are stressed in recent wo rk by, e.g., Paul Feyerabend. Risto Hilpinen has (in another context) argued against this possibility in Rules o f Acceptance and Inductive Logic, but his arguments are certainly not con- clusive if one tolerates different kinds o f acceptance (wh ich seems realistic), and if one introduces a logic that allows the deduction of sound conclusions from such a kind of inconsistent propositions. Although th is lo g ic w i l l d iffe r f ro m Rescher's coherence logic, it follows analogous lines and is surely not more complicated. It would unfortunately go beyond the limits of this paper to go further regarding this point. Anyhow, a l- though an acceptance criterion might lead to the acceptance of " p a rtly" falsified, and even contradictory hypotheses, this would surely be completely unsound for a criterion of truth. An inconsistent set of propositions can never be called true without vast violence to the term. A last difference that I wa n t to mention here is that theories of acceptance differ from theories of truth in their aims and in the status of their results. The conditions f o r accepting a proposition p seem to depend on what we want to use the accepted p for, and on the amount of risk we can permit in this respect. But if 'truth' is meant in a serious sense, one cannot allow that truth would depend on such factors, even if one allows (and has to allow) fa llib ility in his claims to truth. 2. Let us n o w b rie fly have a lo o k at the coherence theory advanced in Rescher's book. A s indicated above, one starts from data, i.e., truth candidates. Each datum can be as complex a proposition as one wants, and the s t of data is lite ra lly the set o f these propositions (and not o f conjuncts o f them). No problem arises if the set is consistent; if one wants to classify the whole set as true, one immediately gets a consistent set of truths (I return to this in the sequel). The importance o f the coherence machinery shows up if the set of data is inconsistent. In this case one has to make a choice: not every member of the set can be classified as true. Before going into more detail, it seems important to note that, according to Rescher's criterion, not only members of the set of data, but also consequences of this set, can be classified as true, even if these consequences NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 397 do not follow from the members of the set that are classified as true (wh a t makes the characteristic non-trivial). This proves the considered coherence crite rio n to be mo re general, and more realistic, than a naive coherence theory would be. Re- turning to the main problem, which consequences of the set of data will be classified as true ? Rescher distinguishes roughly four kinds of consequences of a set o f propositions S. Th e se t o f L-consequences (logical consequences) w i l l contain a ll propositions, because S is in - consistent. Needless to say, this cannot be considered as a set of truths. In order to make clear what the other kinds of consequences are, one has to remember that every inconsistent set S determines a definite number o f " ma xima l consistent subsets" (m.c.s. fo r short), i.e., subsets o f S wh ich are consistent but would immediately become inconsistent if one more member of S was added to them. (Self-) inconsistent members of S will not occur in any rn.c.s.; a ll other members of S will occur in a t least one m.c.s. Rescher ca lls e ve ry .proposition th a t follows fro m each me s. an I-consequence (inevitable consequence) of S, and every proposition that follows fro m at least one m.c.s. a W-consequence (weak consequence) of S. The set of W-consequences will in general (i.e. unless S is consistent) be inconsistent again, whereas the set o f I-consequences will be rather restricted. In general none of these sets can suitably be considered as the set of truths, but the set of truths has to be a subset of the set of W-consequences, and has to contain as a subset the set of 1 -truths a fourth consequence-relation is introduced: P-conse- cquoennces (eprqefuereenticale csonsequence). Suppose o n e h a s so me .device fTo r clasosifying certain m.c.s. as "preferred." The set of dP-coenseqt uenecesr of mS (with respect to this preference distribu- ition)n ise the set of a ll propositions that fo llo w fro m each pre- tferred rn.hc.s. e The con cept of P-consequence is of utmost importance from sa philosopehical point of view. It allows one to express definite trelationsh ips that cannot be expressed b y the p u re ly logical oconcepts o f L f-have to play, at least in my opinion, a central role in epistemo- , I - , a n d W - c o n s e q u e n c e . T h e s e r e l a t i o n s h i p s 398 DIDERIK BATENS logy — and even in an analysis o f what we mean b y 'consequence' in factual situations. I return to the difference between the two kinds of consequence in a later point. Incidentally, it is wo rth noting here that "the wa y the 'content' o f an inconsistent set o f premisses is formulated [ a f f e c t s ] decisively (the 'consequences' o f the set" (p. 102.) This means that there are at least two relativising (and context dependent) factors: the preference distribution, and the wa y the data are formulated. The rationale fo r the first factor is obvious, the rationale fo r the second can be seen best from an example. Suppose each datum is a report of a different observer. I f a part of such a report is disqualified, the whole report becomes questionable: there is a reason to distrust the reporter as to his accuracy. The formulation of the data is then important for identifying propositions that fo rm a whole (each " wh o le " being expressed as, e.g., a conjunctive compound proposition). However, there is a third relativising factor. The same method for introducing a preference distribution is not appropriate in a ll cases. Rescher elaborates some fo u r different methods for preference distribution, and each of these is (more or less) suitable in different situations. It goes beyond the scope of this paper to explore these different methods. Let it only be mentioned that they are a ll worked out ve ry well, and that arbi- trary choices are ruled out as much as possible with in each method. Whereas the relativizations as such seem necessary for a workable truth criterion, it seems unavoidable that a further development o f the theory will require the construction of general rules determining the range o f application o f the different methods, and determining to some extent the methods themselves. One o r more theories, viz., pragmatic ones in my opinion, have to do this job. Needless to say, the fact that these theories are not worked out in the book is not an objection. Rescher himself writes e xp licitly in the "Postscript" th at he considers his book incomplete in certain respects, and that a good deal of further investigation is needed. 3. Some further remarks have to be made about the application of the procedure and its relation with our actual think- ing. Rescher makes a sharp distinction, albeit almost implicitly, NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 390 between, on the one hand, the data, and, on the other hand, the reasons for choosing a definite method for introducing prefer- ences as we ll as the different factors determining the definite preference distribution that one arrives a t with in the chosen method. A positive point in my opinion is that Rescher sees these "reasons" and "factors" as mostly factual in nature. But exactly because he does so, it is amazing to note that he makes such a sharp distinction between these "reasons" and "factors" on the one hand, and the data on the other hand. O f course, some "reasons" and "factors" have to do wit h the sources o f the data rather than with the objects or facts that the data are about and that the truths we are looking for have to be about. But even this does not hold in general. There are at least two problems at hand here. First, can such a sharp distinction be drawn at all ? Next, to what is the truth criterion to apply ? If it applies t o a ll factual truths — th is indeed seems t o b e Rescher's conviction — the next question is whether it has to be applied separately to different aspects o f reality (say, data about sources o f information, and other data) o r to all of them at th e same t ime (this eventually wit h respect t o different problem situations). However, if the truth criterion is applied separately to data about sources and to data produced b y the sources, the situation becomes rather complex. First of all, the former kind of data have their sources themselves, but this is only a minor point. Next, it is impossible to apply the criterion definitively to one set of data before applying it to the other set: the results gained fo r each o f the sets has repercussions on the application o f the criterion to the other set. He re an oscillating process seems the o n ly possible wa y out, always revising the result based on one set after a new result based on the other set is gained. This process ma y eventually come to an end, viz., lead to constant results from some point on. Consider n o w the case wherein the criterion is applied to the whole set of data. The interesting feature here is that no factual material can be used any more to determine which preference method has to be used, and no factual material can be used as an input for the chosen method. A n exception has to be made perhaps f o r the sets o f truths classified as such a t 400 DIDERIK BATENS former points in time, in which case one has to be careful and prevent the o ve ra ll method f ro m becoming static. I t seems interesting to investigate wh a t would happen if we we re to neglect the sets o f propositions classified as true at a former point in time, because this wo u ld reveal wh a t metaphysical, epistemological, or methodological considerations are basically behind th e wh o le procedure. Although th is i s ce rta in ly a n important problem, my proposal will probably lead to a somewhat unrealistic methodology, and perhaps also to inconclusive results; with o u t any doubt one never neglects the set of propositions classified as true at a former point in time. Anyhow, further investigations seem a t hand wit h respect to the data that serve as inputs for the coherence machinery and the data and considerations that (to continue the metaphor) determine the program of the coherence machinery. There is in my opinion an important problem that Rescher does not consider. This problem is most cle a rly seen i f one considers a case wherein the data are consistent. According to the coherence machinery, one ought to classify a ll these data as true. But this is clearly unsound, if one considers indeed all the data without any previous selection. An isolated statement made b y an unreliable source is not like ly to be classified as true. Essentially the same problem arises in cases wherein the data fo rm an inconsistent set: one w i l l wa n t to re je ct even definite I-consequences. Although this is a weakness o f the theory as lite ra lly presented in the book, i t is clear that the only sound way to attack the problem is to do it following the lines of the general approach there, viz., from a wholis tic point of view. One ought not o n ly to consider then a ll m.c.s., but more generally, all "consistent subsets" of the set of data. The notion o f P-consequence can be defined wit h respect to these subsets, in which case we will of course not have any more as a theorem that each I-consequence of S is a P-consequence of S. If this change in the coherence machinery is made, then the notion o f the cohesiveness o f truth, wit h respect to both the content of the set of truths as we ll as the methods that lead to truths (e.g., information about the wa y the data are gathered), would become still more important than it is now in the book. NICHOLAS RESCUER'S COHERENCE THEORY OF TRUTH 401

Rescher writes: "The establishment of a plausibility-grading scheme in terms of the characteristics of these candidates — for example in the case of reports, a reference to the question of how reliable the sources from which they come have been in other instances — can clearly prove both useful and warran- ted." (pp. 111-112.) In general he always presupposes (a) that there is a given number of data that together form a proposi- tional set, and (b) that we have certain information about the data that proves sufficient to apply the coherence machinery to them. However, there are situations wherein these presup- positions seem false o r at least ve ry artificial. First of all, we may be confronted with a set of data that belong to different conceptual structures. In what sense can such data be said to be consistent o r inconsistent ? Th is problem is related t o a more important one that concerns the aforementioned presup- position (b). Is there any way to decide about the truth-status of data without referring to theories, or, i f you wish, to o u r background-knowledge ? Suppose that p is a n I-consequence of some (consistent o r inconsistent) set S, but that p conflicts with a well-established theory. In such a case, it is surely not always rational to classify p as true, and hence to commit one- self to the rejection of the well-established theory. In other cases, data seem to be inconsistent, but are only so in virtu e o f theories. Rescher mentions th e case that " ... a man's sight indicates th a t a st ick h e ld p a rtia lly submerged under wa te r is bent and touch indicates that i t is straight." (p. 301) and resolves the "inconsistency" b y referring to the reliability of the two senses. But sight is in general not less reliable than touch. Furthermore, th e re is n o inconsistency between the stick's looking bent and the stick's feeling straight. So, we need to consider theories — relating h o w something "is" wit h h o w something respectively looks and feels — in order to detect and, needless to say, in order to solve the in - consistency. But even this is not enough. Ho w the matter has to be resolved will largely depend on theories about our senses, light-rays, solids, etc. In short, I want to argue that one has generally to consider the data in relation to theories in order to classify them as true or false. (For an interesting discussion of 402 D1DERIK BATENS part o f this problem, see Ad o lf Grtinbaum's Falsillability and Rationality, to appear in 1973 b y the University of Pittsburgh Press.) The point of vie w I was arguing for is more wholistic than Rescher's, but there is no reason wh y his logic o f coherence could not be used to articulate this point of view more precisely. A ll we have to do is to bring the data and theories together to make them form one inconsistent set. We will then have to assess plausibilities to the data as we ll as to the theories. This application of the coherence machinery does not require that one must be able to give an acceptable account of the whole process of induction in terms o f this machinery. The application is sufficiently useful if it helps to resolve some problems in the short run, and if it helps to reveal problems that have to be resolved b y the modification o f old theories o r b y the invention of new ones. 4. The coherence machinery at issue has to lead from data to truth, viz., about facts that happened in the past and that the data give information about. Rescher also describes how essentially the same machinery can be used fo r inductive purposes. As the results in this case will again have to be truths, it follows that the move will be acceptable only for people who do not share the opinion of e.g., Carnap and Bar-Hillel, that induction leads only to more or less probable propositions, but never to "acceptable" propositions, i.e., leads to different degrees o f rational belief, but never to acceptance. Incidentally, there is another difference between Rescher and La., Carnap in that Carnap starts f ro m absolutely certain evidence (a t least a t some moment) whereas Rescher starts f ro m me re data. Bu t clearly there is no logical connection between these two kinds of difference. The main idea of Rescher's inductive method runs as follows, The different competing hypotheses a re added to the set o f data to form one (of course inconsistent) set S, and the different m.c.s. of S are formulated. The specific problem is now to be solved b y the preference distribution. In general, Rescher pro- poses " to re ly on pattern-conformity considerations" (p. 227), but wit h respect to "general" hypotheses especially, he uses NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 403 probability considerations (applied to the m.c.s. rather than to the hypotheses themselves). Of course, these probability considerations can be regarded as a special case o f pattern-conformity considerations. I t is surely an advantage o f the pro- posed probabilistic procedure that the a p rio ri probabilities of the hypotheses have no ve ry decisive influence o n the final distribution. However, it seems to me that this procedure in - volves some problems wh ich are not mentioned in the book. In cases where an infinite number of hypotheses is considered, there are an infinite number of m.c.s., each of which can hardly receive a higher probability than zero. Perhaps this problem may be solved according to Kemeny's solution of the problem of almost-L-false propositions (" A logical measure function," J.S.L., 18(1953), 289). A second problem seems more difficult to solve. Although the distribution of absolute (a priori) probabilities seems t o b e mo re o r less unimportant f o r Rescher's probability procedure, i t is hard to see h o w the whole me - chanism could ever constitute an adequate "logic of induction" before a ll (translated) problems o f "the adequate m-function" are solved. This becomes clear if one looks at the more general formulation o f th e plausibility procedure: " re ly o n pattern- conformity considerations." There is obviously a n immense number o f pattern-conformities (o r analogies, i f y o u wish ) in every non-simplistic case, given a language that is not aw- fully simplistic and poor. I t seems furthermore completely un ,realistic to suppose that one could take into account all kinds of analogies (which is perhaps a reason for optimism, although there seems to be no solution today as to what analogies have to be considered.) O f course it does not follow from this that Rescher's approach t o induction is misoriented o r has d is- advantages, but it seems to follow that a definitive evaluation of this approach will depend on whether one will ever be able to fill the gaps that one is confronted with now. Anyhow, the application of the coherence machinery for inductive purposes seems to be much harder to evaluate and to specify than its application in the case wherein one wants to construct a set of truths out of a set of mere data about facts that happened in the past. 404 DIDER1K BATENS

5. The main topic o f the book reviewed here is, o f course, epistemology. The most important problems concern the epistemological implications the theory of truth has, and the epistemological considerations the theory is based on. I shall first consider t h e coherentist epistemology a s developed i n th e book, and comment in the next paragraph on the preeminently epistemological problem of the justification of the theory. A ma jo r characteristic o f this theory o f truth is that it is non- foundationalist. One does not gain truths from other truths, but from mere data. Consequently, both protocol sentences and immediate truths from intuition are considered as myths. This characteristic of coherence theories in general is without doubt their major strength in their competition with other theories of truth, and the importance of this book rests in no mean propor- tion on its articulation of a theory of such strength in both an exact and realistic way — the latter in that is that it takes into account a t least some complaints o f history-oriented p h ilo - sophers o f science. The weakness o f a coherence theory lies Obviously in the difficulty t o get a sufficiently close relation between ontic truth and coherentist truth. The essential ro le of data and the specific machinery that leads f ro m data t o truths help to overcome this weakness fo r the specific coherence theory presented in this book to such an extent that one doubts whether another theory could do better (given the kind of epistemic subjects that we th in k ourselves to be). Another characteristic of the coherence theory, viz., that it is wholis tic, is extremely important in this respect. A foundationalist theory, as a very consequence of its nature, will have to develop a criterion that works fo r each particular proposition. The coherentist can do with a general criterion that starts from all data together, o r at least from one set of data related to one problem, and leads to one body o f truths (related b y the ve ry structure of the procedure). From this point of view, the relation between epistemic and ontic truth seems not at all to be a harder problem fo r a coherence theory than fo r a n y other theory of truth. Of course, the foundationalist, and especially the empiricist, will persist in distrusting the coherence machinery, and insist NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 405 on getting "closer to the facts," e.g., in trying to eliminate as far as possible experiential errors. Although Rescher does not mention this point, I can see no reason wh y he would not ap- plaud such a program. There is no reason wh y a coherentist would be "lazy" in looking for data, even if his theory does not especially demand h is increasing the set o f data o r making them more precise. It goes without saying that the increase in the certainty o f experiential techniques is one o f the faiths "we all" have in common. I t is also clear that modern science could not have been what it is now without both the introduc- tion o f experiments, and without the related improvement o f observational (especially measuring) techniques, tha t reduce the role of our senses to more and more simple activities. To be sure, once the well-known "ideal science" will be there, all coherentist machinery w i l l b e superfluous. B u t a s t h is u n - fortunately is not going to happen before the end of time, we are confronted in the meantime with at least two problems. The first problem is that we have to apply some criterion to the inconsistent sets o f data that we still come upon. The second problem is that, given where we are now, any improvement has to start from some conjectures, and it would be comforting if these conjectures we re a t least p a rtia lly true. Both the e m- piricist and th e coherentist w i l l use information about th e liability of the different experiential reports, but thanks to the wholistic character o f the coherence theory, th e coherentist will n o t b e obliged t o a p p ly h is crite rio n t o each d a tu m separately. I thought the above (rather dynamic) considerations worth mentioning because one might misinterpret the fact that Rescher's book concentrates upon the static description, and conclude f ro m this that i t disregards th e importance o f im- proving experiential techniques, and even of increasing factual information (Rudolf Carnap's Logical Foundations o f Prob- ability was misinterpreted in the same manner). Apart f ro m b e in g non-foundationalist a n d wh o listic, t h e coherence th e o ry under consideration is fallibilistic. a l l claims to factual knowledge can never be made categorically and definitely, b u t o n ly conditionally and provisionally" (p. 325). I mentioned this at the ve ry beginning, and I cannot see 406 DIDERIK BATENS how one could object to this without making the class of known truths almost empty. Another characteristic of the coherentist epistemology is what Rescher calls the "dualism of contingent knowledge": knowledge of something as a datum and kn o wl- edge o f something as a fact. Perhaps " p lu ra lism o f factual knowledge" wo u ld even be preferable, f o r it does n o t seem clear to me that no other kind of knowledge is to be introduced. It seems plausible, fo r instance, to attribute some knowledge status t o (perhaps mu ltip le kin d s o f) accepted propositions, even if these cannot be considered as true. A further study of the relation between acceptance and truth would perhaps be instructive. I t seems indeed advisable to distinguish different kinds of acceptance (and different criteria) with respect to the purpose the accepted propositions have to serve. I t seems not advisable to extend the same mu ltip licity to truth — to have mutually inconsistent sets o f truths seems to be a contradic- tion — but a comparative study could help to articulate more precisely the different factors that play a role in the plausibility procedure. 6. Rescher presents an interesting chapter on the problem of validating a th e o ry o f tru th i n general, a n d discusses t h e different alternatives before us. He comes to the conclusion that one has to turn to pragmatic considerations at this point. Just as correspondence considerations also play a role in the coherence theory under the form of data, pragmatic considerations p la y a ro le i n the justification o f the theory. A s th e coherence theory offers a criterion, i.e., a method, th e " a p - propriate mode of validation is to show that it works" (p. 232). This, o f course, cannot be done better by anything other than pragmatic considerations. The main difference with a pragmatic theory of truth is that Rescher applies the pragmatic considerations for validating the whole method, and not for validating separate claims to truth. This pragmatic turn saves the coherence theory from an a p rio ri metaphysics, and stresses in the same manner as did the correspondence turn wit h respect to the data, the importance of factual and practical considerations. The problem is of course what we are going to use as information fo r judging whether the method wo rks o r not. One can NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 407 clearly not validate the criterion in saying that i t wo rks in that it leads to truths. This would be plainly circular. To escape circularity, Rescher turns again to the data. These are, to be sure, n o t true, b u t th e y h a ve nevertheless some cognitive status (cf. supra). The argument has then to re ly on the fact that the method delivers us a vast ma jo rity of the most plausible data as truths. It seems to me that a stronger and more plausible validation of some criterion of truth (or some method in general) could be reached if attention is paid to the facts (a) that there is no need for accepting one criterion once and f o r all, and (b) th a t a criterion is not applied at one point in time but during some interval of time. In applying a truth criterion more than once and starting from different sets of data, one will be forced to consider previous decisions about the truth of certain propositions as mistakes. A criterion that cannot lead to mistakes must be rejected a priori for reasons that will become clear later on. The fact that a criterion leads to mistakes provides some kind of information of its accuracy in the short run, judged " b y its own standards." I f one may suppose that there is some accuracy in the results o f the criterion (and note that there wa s surely some accuracy already in pre-scientific thought — man- kind did among other things not die out), and if one may suppose that later applications lead to more trustworthy results than former ones (La., because they deal with more data), one has reasons to substitute the used criterion for one that would have made less mistakes. The fact that one is able to perform this kin d of "counterfactual" reasoning about criteria, entails that, if one is able to invent at least variations on the method he uses, then he ma y be able to substitute some criterion b y "its" (in the strong sense) improvement — and I think indeed that a n y improvement o f o u r methods proceeds along these lines. Th e problem whether and h o w one is able t o in ve n t improvements of methods is less important here, as the main aim of this argument has to do with the justification o f some method once it is invented. One may of course doubt whether such improvements occur o n th e general le ve l o f the tru th criterion, or rather on the lower level of measuring techniques, 408 DIDERIK BATENS etc. However, there seem t o me t o b e various reasons f o r building in this dynamic on the most general level. The reasons for doing so wit h respect t o Rescher's t ru t h crite rio n a re among others the following: (a) there would be no need for a truth criterion at all if our observations were completely accur- ate, a supposition from which people probably started original- ly, (b) some of the alternative methods (to be applied in different contexts) that Rescher offers are certainly not used in that exact and sophisticated manner today, (c) t h e invention o f such a n alternative does ce rta in ly n o t determine it s e xa ct field o f application, and (d) t o consider such a n alternative makes sense o n ly i f fo rme rly used methods we re a t least to some extent less satisfactory. A more general reason f o r building in this dynamic on the most general level is that it seems to be the best warrant for the reliability of the validating argument for any theory of truth.

The above comments are surely not an actual validation of Rescher's coherence th e o ry w i t h respect t o contemporary science. They have to do with a metatheory that has to guide changes in the used method, viz., the truth criterion. The validation o f this criterion rests on the ve ry fact that it is guided by such a metamethod. Also Rescher's arguments have to do with a way of validating a criterion rather than that they constitute an actual validation; the latter wo u ld indeed require to su m u p factual information about th e application o f the criterion o f truth. But if we are concerned wit h a method o f validation of a criterion of truth, it seems to me that we have to make it rest on dynamic considerations, and that it is o n ly too optimistic to believe that we could validate a criterion at one point in time, stopping as it were to apply the criterion for a short time, and reflecting upon the fact that i t is a good one. I t seems furthermore clear to me that we shall have to improve a criterion many times; and we need a method to guide us here. I f this is so, then the validation o f any criterion o f truth has to rest above a ll on the characteristics o f the metatheory that determines the changes in the criterion. I t should be noted incidentally, that we need a "self-correcting" criterion NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 409 for truth if we want to be able to "correct" the criterion itself by a metatheory. Although no sufficient validation o f Rescher's approach to truth has been given, I hope to have made clear that this kind of approach has to be followed up and elaborated if we ever want to come to a sound theory o f epistemic truth. 7. In this last point, I want first of all to say something about a ve ry general difference between this book and many other works on epistemology, methodology, and logic. Rescher takes into account and builds in in various parts of his theory factors that I shall label in general as contextual. Consider first o f all the logic o f coherence. I n standard kinds o f logic we are told what can be derived from some set of propositions; once these are given as premisses, the consequences are determined in an absolute wa y irrespective o f its purpose. Even if some logicians are so kin d as to a llo w different logics to express different kinds of consequence relations, most of them will not allow one and the same consequence relation to be different according to the context. However, in Rescher's logic of coherence e xa ctly such a kin d o f diversification exists: different preference distributions will determine different sets o f P-consequences o f one and the same (inconsistent) set o f propositions. This gives the logic a special attractiveness and allows for th e p o ssib ility o f dealing w i t h non-simplistic cases. I t enables us to give to the different premisses a different status with respect to the derivations that can be made fro m them. Hence, we are not only able to take into account the presup- posed and unique tru th value o f a proposition, b u t also it s "epistemological status." N o t o n ly in it s application i n th e theory of truth, but also in its application to a large number of other cases o f "reasoning fro m inconsistent premisses" (a whole chapter of the book is devoted to such cases) does such a lo g ic seem t o b e th e o n ly w a y t o g e t sound results i n systematizing o u r actual cognitive behavior and explicating the concepts we have in mind. Analogous characteristics can be taken notice of in the construction of the methodological part of the theory. There is a variety of specific methods, each of which is to be applied in 410 DIDERIK BATENS the situation wherein it is suitable, and although there is certainly wo rk to be done about the delineation of such methods, there is n o reason f o r objecting t o the mu ltip licity o f such methods. Wh a t is determinative in a ll these cases is the context (in th e broadest sense) o f the application o f th e tru th criterion. He re again, th is contextual relativization gives a t least hope f o r a sound result fro m the point o f vie w o f o u r actual behavior and of the concepts we have in mind, however vague they might be. One has only to compare these characteristics wit h a b o o k as excellent as Rudolf Carnap's Logical Foundations of Probability to see the striking difference. To be sure, th is contextual relativization makes the exact articulation of the theory more complex and difficult, b u t at least it saves one from the lures of a too elegant theory, wherein the disadvantages o f its ve ry simp licity are a ll too easily over- looked. Let it be mentioned that Rescher's book contains a lo t o f interesting and we ll developed fo rma l theories, such as th e logic o f coherence, an orthodox non-classical theory o f truth (in it s relation wit h three-valued logic), a fo rma l th e o ry o f plausibilities, etc. A ll these do not only help to articulate the coherence theory of truth in an exact way, but are interesting for their own sake. Furthermore, the wide range of application of the coherence machinery as brought forth in the book opens interesting perspectives on numerous philosophical problems. A special chapter is devoted to the applications of the coherence analysis i n information processing. These applications include, among other things, determining written passages of a book, spoken utterances, and group consensus reports — a ll in the face of conflicting information — and collating the results of different and conflicting sensory reports (but compare point 3.). I n a ll such cases a rather clear distinction can be drawn between data about the "facts" a n d data about th e "sources," and the problems I mentioned before in this respect do consequently not arise here. The application of the coherence machinery t o such problems leads t o plausible results and enlightening insights. A l l this stresses the power o f the coherence machinery. NICHOLAS RESCHER'S COHERENCE THEORY OF TRUTH 411

In concluding this essay, i t should be mentioned that this book of Rescher's combines a real philosophical point of vie w with a realistic perspective on the actual development of our scientific knowledge. I t is writte n in a clear language (some- times perhaps a bit too redundant) without violating the difficult epistemological problems o f the subject. I hope that this paper shows the importance o f the philosophical theory presented in this book and of its further development, and that it suggests some tentative roads f o r such development.

Rijksuniversiteit Gent Diderik BATENS

I am indebted to Professor Nicholas RESCHER for interesting and clarifying discussions of his book and to Mr. Brian BAKER for his comments on an earlier draft of this paper. This paper was written during a stay at the University of Pittsburgh, sup- ported b y a Fellowship o f the Belgian American Educational Foundation.